Application of CDCARLA Technique in Designing Takagi-Sugeno Fuzzy Logic Power System Stabilizer (PSS) M. Kashki, A. Gharaveisi, and F. Kharaman Abstract. One of the effective methods for damping oscillation due to occurring disturbances in power systems and keeping synchronism of machine is using power systems stabilizers (PSS) in excitation circuits of synchronous generators. In past decades, designing and tuning of PSS have been interested for many researches and scientist, because inaccurate setting of PSS not only cause damping of oscillations but also may help to amplify instability and lead to loss of synchronism. In this paper, Takagi- Sugeno Fuzzy logic PSS designed with novel method call CDCARLA have been proposed for single machine-infinite bus (SMIB) power system model and compared with two IEEE standard PSS. Simulation results shows that designed PSS has better performance and more robustness than other PSSs. Brilliant properties of CDCARLA method is that it does not require to system dynamics and any other information about power system. In addition, this method does not ignore any nonlinear feature of power system; consequently, CDCARLA method can be one of automatic design method. Index terms: Power system stabilizer, PSS, Reinforcement learning automata, SMIB, Transient stability. I. INTRODUCTION UE to increasing complexity of electrical power systems, there has been increasing interest in the stabilization of such systems. Power system stabilizers are one the most effective devices for stabilizing and damping of low frequency oscillations and as a result increasing stability margin of power systems [1]. PSS prepare supplementary input signal in-phase with synchronous rotor speed deviations to excitation systems and cause generator stability. In the last two decades, various types of PSS have been designed. Conventional power system stabilizers (CPSS) are one of the premiere PSS that is composed of fixed lag-lead compensators and widely used in power systems [2]. Adaptive controller based PSS have been used in many applications [3-4]; however, most of these controllers are based on system identification and parameter estimation; therefore time consuming computations are required. Fuzzy logic based PSS (FLPSS) have been interested recently in many applications [5-8]. Low computation burden, simplicity and robustness make them suitable for stabilization purposes. Different methods for designing fuzzy based PSS proposed so far such as genetic algorithm [9] and neural networks [10]. In this paper, a novel designing method of fuzzy controllers proposed and a typical Takagi-Sugeno Fuzzy logic PSS was designed using this method. The proposed method is based on Reinforcement Learning Automata (RLA) and includes two stages; in first stage, best variation limits of controller parameters (fuzzy rules coefficient) are obtained using DARLA 1 , and in second stage the best value of these parameters in specified limit is determined. Second stage firstly proposed by Howell et. al. [11] for a vehicle suspension 1 Discrete action reinforcement learning automata control application in 1997 and named CARLA 2 . CARLA has numerous advantages such as high speed convergence; however, it requires pre-specified decision variables variation limits. These limits can be obtained using any simple method such as local linearization. Therefore, CARLA method requires system dynamics in initial step and this property count as disadvantage; moreover, these system equations can not be achieved easily for any application. DARLA has similar properties due CARLA such as fast convergence, system dynamics independency, considering fully nonlinear characteristic and so on. Consequently, as indicated by proposed method by combining DARLA and CARLA methods we can approach to an automatic designing structure of PSS. This proposed method called CDCARLA 3 and has all of above worthwhile properties. For evaluating performance of CDCARLA designing method, stabilization behavior of designed Takagi-Sugeno Fuzzy PSS simulated and compared with two IEEE standard PSS (PSS2B and PSS4B) for single machine infinite bus power system model. The simulation results shows that proposed method have a better performance against other PSS for various power system electromechanical disturbances. The structure of paper is as follows: in section II Takagi- Sugeno fuzzy PSS structure presented, section III describe power system model, in section IV the novel design methodology has been described, section V shows the simulation results of proposed design method, and section VI conclude the paper. II. POWER SYSTEM STABILIZER In this paper, a Takagi-Sugeno Fuzzy Logic PSS (TSFLPSS) is designed. This TSFLPSS has two input variables: rotor speed deviation ( ω ∆ ), and accelerating power ( e m a P P P − = ∆ ). For both input variables, three triangular membership functions have been considered as shown in Fig. 1. -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 ∆ω Membership Value MF 1,1 MF 1,2 MF 1,3 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 ∆Pa Membership Value MF 2,1 MF 2,2 MF 2,3 Fig 1. Membership functions of input variables 2 Continuous action reinforcement learning automata 3 Combinatorial discrete and continuous action reinforcement learning automata D